One diplomat's Monday

Follow her morning with social media monitoring AI tools.

A simulated world. The tools really ran; this replays what they did.

It is the morning of Monday 10 November in our fictional embassy. A diplomat wants to learn what social media posts reveal about an international dispute she is following. Scroll to see the questions she wants answered, how her AI tools work and what the tools send back as an answer. The backdrop is a dispute between two countries – Ravelia and Taldara – over a shared ore field. The treaty governing the co-run mining operation is due for renewal. Low-level fighting erupted several weeks ago. A third round of talks to resolve the dispute will occur in 4 days' time. You can read more about our simulated world and how we built it here.

08:30

To: AI Social Media Monitor

What happened in last week's social media discussion of the dispute?

Used to take 6–8 hoursNow takes 25 min with AI

  1. The AI…
  2. read last week's 1,246 posts from 39 channels
  3. grouped the channels by who they speak for
  4. tracked 5 pre-defined topics related to the dispute
  5. compared last week with the two months before
  6. spotted 13 unusual changes in what groups wrote about
  7. read the stance of every post on the 4 topics with unusual changes
  8. spotted 9 shifts in group stance
  9. wrote up each change, with quotes where it had them
  10. Passed to the AI analyst
  11. checked each change against the group's history and recent events
  12. picked the 5 that matter, and asked why each happened
  13. checked its own claims against the sources
  14. concluded government-friendly channels started to argue for a deal

or continue scrolling.

08:55

To: AI Video Translator

The Ravelian government spokesperson's video is circulating. What did he actually say?

Used to take 1.5–2 hoursNow takes 12 min with AI

  1. The AI…
  2. found the video and its transcript
  3. translated it, flagging 4 unclear passages
  4. pulled up his government's recent statements
  5. compared today's wording with the usual
  6. spotted 1 word omitted from a standard demand, twice
  7. picked out the lines that matter
  8. concluded he called the call-up a direct threat, and omitted "unconditional"

or continue scrolling.

09:07

To: AI Note Writer

It looks like the Ravelian government is preparing its supporters for a softer negotiating stance. Draft a note for me on the social media evidence for that.

Used to take 2–3 hoursNow takes 13 min with AI

  1. The AI…
  2. gathered the morning's results and the posts they cite
  3. listed the evidence for her view, and against it
  4. weighed how far each source can be trusted
  5. drafted the note in her ministry's house style
  6. checked its draft against the sources, as a separate reader
  7. revised the draft once
  8. concluded the channels shifted, but a softer stance is unproven

or continue scrolling.

09:20

To: AI Note Checker Council

Check the note (challenge, check and cite).

Used to take 1–2 hoursNow takes 17 min with AI

  1. The AI…
  2. listed the note's 8 key claims and found their sources
  3. traced each source back to where it started
  4. sent it to three council members, who checked it separatelyFact-checkerChallengerStrategist
  5. questioned the second omission of "unconditional": the line at 02:40 is garbled
  6. brought their views together, with a source for every point
  7. concluded a softer stance is still unproven, and asked for two corrections

or continue scrolling.

More than a day's work, done in about an hour, with her reviewing each result.

10–15 hused to take
About 1 hnow takes with AI

Risks, and how they're handled

It can misread. Every result shows its sources, and she checks before anything goes out.

It can overreach. A council of three AI checkers reads each note separately, and she acts on what all three agree on and weighs the rest herself.

Loud stories drown quiet ones. The monitor compares each group with its own past, not with the week's noise.

Sources can echo. The review traces a repeated claim back to where it started.

How this was made
  • A simulated world. About 11,000 channel posts, news items, statements and video transcripts, all invented and written by AI for this demonstration. Every state, person and outlet is invented.
  • The tools really ran. Stages 1 to 3 replay real runs from November 2026: scripts did the finding and counting, and an AI model did the reading and writing. For stage 4, the council was run again with the tool's own instructions and the full text of every source, because the first run had lost the text of three sources. Her requests and checks were scripted in advance, so the tools ran without a person stepping in. One section of the note, "Who is behind the slogan", was written by the team from the same sources. The reports shown here are condensed and edited from what the tools wrote.
  • Built on real work. The social media monitor extends Artem Apostolov's discourse monitor and his method, applied here to an invented conflict.

Now takes: estimated from the tools' recorded run times, with their tasks run side by side, plus her checking. Used to take: the AI for Diplomacy team's estimate.